Explainable machine learning–based decision support for predicting heavy metal adsorption by diatomite

Heavy metal (HM) contamination poses a serious threat to environmental and human health, making effective removal of HM ions an important environmental challenge. Diatomite is a natural, low-cost adsorbent with demonstrated potential for HM removal. Machine learning (ML) provides a flexible approach for predicting adsorption under multivariable experimental conditions. This study presents an explainable ML framework for predicting the HM adsorption capacity of natural diatomite under different experimental conditions. A consolidated dataset of 636 unique adsorption observations for Cd 2+ , Pb 2+ , Cu 2+ , and Zn 2+ was used to evaluate six ML models – decision tree, random forest, extreme gradient boosting, histogram-based gradient boosting machine, support vector regression (SVR), and neural network (NN)—using repeated nested cross-validation. The NN achieved the lowest 𝑅 ⁢ 𝑀 ⁢ 𝑆 ⁢ 𝐸 (0.5995 mg/g) and the highest 𝑅 2 (0.9975), although its 𝑅 ⁢ 𝑀 ⁢ 𝑆 ⁢ 𝐸 did not differ significantly from that of SVR after dependence-corrected pairwise testing with Holm adjustment. SHapley Additive exPlanations (SHAP) analysis identified initial concentration, metal identity, and electrolyte cation charge as the strongest contributors to model predictions. The selected NN was additionally deployed in a Windows-based graphical user interface (GUI) for real-time prediction under user-defined experimental conditions. This study demonstrates the potential of intelligent models as practical tools for predicting the adsorption of HM ions on diatomite in environmental remediation applications.